Start with the caveat, because it outranks every number below. Range Breakout EA does not reproduce. Our tester journal for the audited run counts 2,481 randomization prints, so our catalog flags this expert non deterministic. We did not infer that from the label. We measured it. Two short runs on a byte identical settings file and a byte identical tester configuration, on the same machine, six minutes apart, over 2018-01-01 to 2018-06-30, produced 175 deal rows each and 174 of them differed.
What differed is as interesting as the fact that something did. Deal type, direction, volume, symbol, order number and sequence were identical across all 175 rows. Time, price, profit and balance were not. The two runs netted -$15.94 and -$21.52 at profit factors of 0.93 and 0.91. The z score came out at -0.20 in one run and +0.22 in the other, so even the direction of that statistic was drawn rather than computed. The EA writes a randomization number into the journal for every trading day and the two runs drew different ones.
The tester is not the source of the spread. Both runs generated the same tick stream down to the last tick, 20,607,743 on XAUUSD in both, and the same for all seven symbols involved. The line is therefore clean. Structure reproduces, money is drawn. Everything below describes one draw of the long run.
On a 100,000 USD measurement frame at 1:5000 with fixed 0.01 lots, the EA took 1,525 trades across 3,050 deals and finished at +$2,128.44 after our commission retrofit, for a final balance of $102,128.44. The tester head prints $2,250.44 before that retrofit. Gross profit before costs was $2,250.65, commission took -$122.00 at $3.50 per lot per side across all 3,050 deals, and swap took -$0.21 across 19 overnight trades. Costs are 5.4% of gross, which our engine rates ok.
The reconstruction is tight. Pairing came from the tester journal with a coverage of 1.0 and a confidence of exact, 0 pairing violations, 0 chronology violations and 0 duplicate deal ids. The reconstructed final balance matches the report to the cent. The journal also records 0 failed entries and 0 failed modifies, which is rarer in our catalog than it sounds.
One data quality caution stands. The EA takes its first trade on 2017-01-03 08:35:22, two days into a window that opens on 2017-01-01, which leaves 9.65 effective years. Bar coverage of the window is 97.92%. Neither is alarming. Both are stated because the window is ours, not the vendor's.
The EA was active on 1,507 trading days out of 2,520 in the window, an active share of 59.8%. Eighty percent of the profit was made on 21 of those active days, which is 1.4% of them. Our concentration dimension rates that red.
The best day in the run is 2026-03-03 at +$235.84, which is on its own more than a tenth of the whole result. The worst is 2026-03-16 at -$54.15. Remove the twenty best trades and the run keeps +$470.24 of its $2,128.44, because those twenty carry $1,658.20 between them. Remove the ten best and $1,104.35 remains. That is what a thin, clustered edge looks like when you poke it.
There is no martingale here. Position size after a loss goes up in 0.0% of cases and the median size ratio after a loss is 1.0. The EA holds at most 2 positions at once on a single symbol, stacks entries in 0.6% of cases and held both directions at once in 9 moments across the whole run. Our structure dimension flags those two as info, not as risk.
All 1,525 trades belong to one setup group. 780 were long and 745 short, the win rate is 51.5%, the median holding time is 11.38 hours and the busiest entry hours are 7, 8 and 9. Exits split into 193 stop losses and 1,332 signal or time exits, and 0 stop exits closed in profit, so there is no trailing stop signature. This is a plain intraday breakout system that does what its name says.
It also survived. Across 3,050 simulated start points the share of runs that died is 0.0%, and the deepest end of day drawdown in the whole run is 0.49%, a $500.41 dip from 2025-11-17 to 2026-02-12 that recovered on 2026-04-21 after 111 trading days underwater. The longest episode of the 66 we counted ran from 2021-11-09 to 2023-12-19, 550 trading days underwater for a depth of $247.95. Shallow and slow, not deep and fast.
Over the tested span gold moved from 1,183.572 to 4,319.824, a rise of 264.98%. Holding 0.01 lots long through the same window and doing nothing else would have produced $3,136.25. The EA produced $2,128.44 with 1,525 trades. Its long trades made $1,379.12 and its short trades $749.32, so the shorts did contribute, which is not the case for every gold expert we have measured.
The yearly line is calm. There is exactly 1 losing year, 2019 at -$17.83, and the two strongest are 2020 at +$377.64 and 2026 at +$811.91 from only 93 trades. That 2026 figure needs the mechanical footnote that applies to every fixed lot backtest on gold. At a constant 0.01 lots the same percentage swing is worth many times more dollars at the end of a window in which the underlying rose 264.98% than at the start. The end of day view puts the whole run at a CAGR of 0.21% and an end of day Sharpe of 0.94 against the 1.84 the report head prints.
The break even shock, the extra cost per trade that takes the run to zero, is $1.40. At $0.50 extra per trade the run still nets +$1,365.94. At $1.00 it nets +$603.44. At $2.00 it becomes -$921.56. The average position is 0.01 lots, so that entire margin lives inside a spread and slippage assumption.
Our resampling agrees that the sign is probably real and the size is not pinned down. Over 1,000 paths the probability of a net loss is 0.003, the profit factor lands between 1.109 and 1.448 at the 5th and 95th percentile with a median of 1.265, and expectancy per trade runs from $0.61 to $2.22 with a median of $1.38. Median time under water across those paths is 443 days and the 95th percentile is 1,016 days.
Against IQ Capital Classic our sweep records 0 account deaths at every size from full down to one eighth and 0 breaches of the per position loss limit. The modelled floor breach probability over 1,000 paths is 0.0 at every sizing. The floor is simply not the problem for this expert.
The consistency rule is. It caps the best day at 30% of annual profit and this profile breaches it in 4 years. In 2018 the best day was $26.13 against a year profit of $17.16, a share of 152.3%. An EA that earns on 1.4% of its active days will keep doing that to a consistency rule. The withdrawal rate the prop sweep models at full size is 0.21% per year, against the 0.22% our own monthly replay paid. Our prop fit verdict is red for the consistency reason and not for a survival reason.
The product reached the market on 2024-08-25 and was last updated on 2026-08-23, with 30 recorded updates in between. Of the tested history, 1,235 trades and $992.74 at a profit factor of 1.20 lie before the release. The window from release to last update holds 290 trades and $1,135.70 at a profit factor of 1.37. The window after the last vendor update, the only slice that is out of sample by construction, holds 0 trades, because the last trade of the run closed on 2026-08-13, ten days before that update landed. There is no out of sample evidence here yet. The listed price on 2026-09-02 was 649 USD.